Overview Figures

Let’s start by looking at the site distribution. We can color the sites based on their designated category (urban vs rural). We will also highlight Wilderness Park due to collections that happen there for the activity monitoring of Agelenopsis pennsylvanica. We will also fill the site labels of the sites where A. pennsylvanica are collected for the lab choice test.

Sites are randomly distributed across Lincoln, Nebraska (urban) and into the surrounding rural area (rural).

Let’s now take a look at a summary of the vibratory noise recorded across the experiment. Here, we order sites by the lowest average Leq (left) to the highest average Leq (right). Sites are divided along the horizontal axis and further divided by visit number. Microphones are divided across the vertical axis, further separated by substrate and hour. Each box represents the average Leq of an hour. Missing bars indicate failed recording trials (i.e., the microphone fell during recording or received damage from wildlife).

Spatial Variation in Vibratory Noise

Does vibratory noise vary over space? Let’s first look at average Leq across sites. We will use all 23 sites.

Site averages vary from -55 dB to -70 dB (~15 dB difference). The sites with the highest Leq appear to occur near highly traveled roads (i.e., highways/interstates).

Let’s test this hypothesis by using the PCA analysis that reduced measures of potential traffic impact and see if that predicts vibratory noise levels.

Let’s first take a look at the raw data:

We performed lmer ( mean_leq ~ Dim.1 * Substrate + (1 | Site) , data = dayavgl ) overall and for category subsets. We also tested each with rlmer. If lmer gave the same results as rlmer, we used lmer. Otherwise, we used rlmer. We found:

## [1] "lmer results"
Estimate Std..Error df t.value Pr…t..
(Intercept) -62.199 0.6942 21.04 -89.595 0.0000
CategoryRural -5.801 1.3588 21.01 -4.269 0.0003
Estimate Std..Error df t.value Pr…t..
(Intercept) -62.9753 0.5173 23.75 -121.7412 0.0000
Dim.1 1.7958 0.3067 23.62 5.8561 0.0000
SubstratePlant -1.3155 0.2228 271.07 -5.9037 0.0000
Dim.1:SubstratePlant -0.0906 0.1311 271.25 -0.6914 0.4899
Estimate Std..Error df t.value Pr…t..
(Intercept) -65.6890 1.1709 17.49 -56.0994 0.0000
Dim.1 4.4971 1.0832 17.39 4.1517 0.0006
SubstratePlant -1.7511 0.5452 198.73 -3.2121 0.0015
Dim.1:SubstratePlant 0.2863 0.4994 198.69 0.5732 0.5671
Estimate Std..Error df t.value Pr…t..
(Intercept) -69.7239 4.4612 4.995 -15.6289 0.0000
Dim.1 -0.7928 1.6125 4.899 -0.4917 0.6442
SubstratePlant 4.5194 2.6552 70.374 1.7021 0.0932
Dim.1:SubstratePlant 1.9635 0.9538 70.273 2.0585 0.0433
Estimate Std..Error df t.value Pr…t..
(Intercept) -61.378 0.5837 20.94 -105.161 0.0000
Dim.2 2.169 0.6258 19.21 3.466 0.0026
SubstratePlant -1.326 0.2226 271.99 -5.956 0.0000
CategoryRural -5.773 1.1275 19.07 -5.120 0.0001
Dim.2:CategoryRural -2.912 1.5809 19.15 -1.842 0.0810 
## [1] "rlmer results"
Estimate Std..Error t.value P
(Intercept) -62.846 0.5641 -111.403 0
CategoryRural -5.472 1.1042 -4.955 0
Estimate Std..Error t.value P
(Intercept) -63.4368 0.4571 -138.7883 0.000
Dim.1 1.7401 0.2710 6.4212 0.000
SubstratePlant -1.1343 0.1895 -5.9856 0.000
Dim.1:SubstratePlant -0.1041 0.1115 -0.9342 0.351
Estimate Std..Error t.value P
(Intercept) -65.8914 1.0922 -60.3286 0.0000
Dim.1 4.5827 1.0104 4.5357 0.0000
SubstratePlant -1.5987 0.5104 -3.1324 0.0020
Dim.1:SubstratePlant 0.3185 0.4675 0.6812 0.4965
Estimate Std..Error t.value P
(Intercept) -68.3135 4.0176 -17.0034 0.0000
Dim.1 -0.1184 1.4560 -0.0813 0.9354
SubstratePlant 1.6596 1.6504 1.0056 0.3179
Dim.1:SubstratePlant 0.8699 0.5927 1.4676 0.1464
Estimate Std..Error t.value P
(Intercept) -61.783 0.5623 -109.876 0.0000
Dim.2 2.042 0.6058 3.371 0.0008
SubstratePlant -1.137 0.1898 -5.991 0.0000
CategoryRural -5.820 1.0919 -5.330 0.0000
Dim.2:CategoryRural -2.606 1.5306 -1.702 0.0897

Let’s check the assumptions:

Let’s graph the results:

By Category - lmer

Urban sites had louder daily vibrations than rural sites (t = -4.27, df = 2, 295, P < 0.001).

Overall PC1 - lmer

Daily average Leq has a significant positive relationship with Principal Component 1 - road vibratory noise (t = 5.86, df = 4, 295, P < 0.001, cond R\(^2\) = 0.81, marg R\(^2\) = 0.51). Daily average Leq was significantly higher on manmade material than plant material (t = -5.9, df = 4, 295, P < 0.001). There is no interaction (t = -0.69, df = 4, 295, P = 0.49).

Urban Subset - lmer

Daily average Leq has a significant positive relationship with Principal Component 1 - road vibratory noise (t = 4.15, df = 4, 217, P = 0.001, cond R\(^2\) = 0.81, marg R\(^2\) = 0.51). Daily average Leq was significantly higher on manmade material than plant material (t = -3.21, df = 4, 217, P = 0.002). There is no interaction (t = 0.57, df = 4, 217, P = 0.567).

Rural Subset - rlmer

There was no sigificant correlation between daily average Leq and Principal Component 1 - road vibratory noise (t = -0.08, df = 4, 78, P = 0.935, cond R\(^2\) = 0.64, marg R\(^2\) = 0.05), substrate (t = 1.01, df = 4, 78, P = 0.318), or interaction (t = 1.47, df = 4, 78, P = 0.146).

Overall PC2 - lmer

Daily average Leq has a significant positive relationship with Principal Component 2 - road vibratory noise (t = 3.47, df = 5, 295, P = 0.003, cond R\(^2\) = 0.81, marg R\(^2\) = 0.53). Daily average Leq was significantly higher on manmade material than plant material (t = -5.96, df = 5, 295, P < 0.001). Urban sites had louder vibrations than rural sites (t = -5.12, df = 5, 295, P < 0.001). There is a trend of an interaction between PC2 and Category (t = -1.84, df = 5, 295, P = 0.081).

Let’s take a closer look at the substrate.

Manmade structures - Paneling, Metal, Concrete, Brick, Wood

Plant structures - Herb, Tree, Shrub, Vine

Brick, paneling, and shrubs carried the highest amplitude vibrations Bricks and herbs have the steepest slopes, which might suggest these substrates are affected by vibratory noise. Wood in quiet areas have high vibrations, probably as a result of people and pets walking on porches.

Temporal Variation in Vibratory Noise

Season

It seems like the rural sites got louder on the third visit.

We performed lmer ( mean_leq ~ Visit * Category + (1 | Site) , data = dayavgl_20 ) and as a rlmer and found:

F Df Df.res Pr(>F)
Visit 2.454 3 241.61 0.0639
Category 20.493 1 18.97 0.0002
Visit:Category 1.164 3 241.45 0.3241 
## $`emmeans of Visit`
##  Visit emmean    SE   df lower.CL upper.CL
##  1      -65.5 0.650 27.1    -66.8    -64.2
##  2      -65.8 0.634 24.5    -67.1    -64.5
##  3      -64.6 0.643 25.9    -65.9    -63.3
##  4      -65.2 0.631 24.0    -66.5    -63.9
## 
## Results are averaged over the levels of: Category 
## Degrees-of-freedom method: kenward-roger 
## Confidence level used: 0.95 
## 
## $`pairwise differences of Visit`
##  1               estimate    SE  df t.ratio p.value
##  Visit1 - Visit2    0.279 0.392 242   0.711  0.8927
##  Visit1 - Visit3   -0.867 0.407 242  -2.129  0.1468
##  Visit1 - Visit4   -0.260 0.388 242  -0.671  0.9081
##  Visit2 - Visit3   -1.146 0.382 241  -3.002  0.0156
##  Visit2 - Visit4   -0.539 0.360 241  -1.498  0.4401
##  Visit3 - Visit4    0.607 0.376 241   1.614  0.3728
## 
## Results are averaged over the levels of: Category 
## Degrees-of-freedom method: kenward-roger 
## P value adjustment: tukey method for comparing a family of 4 estimates
Estimate Std..Error t.value P
(Intercept) -67.4704 0.9255 -72.9017 0.0000
Visit1 -1.1953 0.5640 -2.1192 0.0350
Visit2 -1.1350 0.5292 -2.1446 0.0329
Visit4 -0.7818 0.5303 -1.4743 0.1416
CategoryUrban 4.8348 1.0907 4.4329 0.0000
Visit1:CategoryUrban 0.7603 0.6685 1.1373 0.2564
Visit2:CategoryUrban 0.3021 0.6267 0.4820 0.6302
Visit4:CategoryUrban 0.2537 0.6177 0.4107 0.6817

Let’s check the assumptions:

Let’s graph the results:

To investigate whether noise varied across the season, we used a linear mixed model with visit number and category, and their interaction with site as a random factor. There was a trend that daily average Leq varied across the 2020 season (F = 2.45, df = 8, 268, P = 0.064, cond R\(^2\) = 0.75, marg R\(^2\) = 0.39). A post hoc test suggests that visit 3 was significantly louder than visit 2 (t = -3.002, P = 0.016). Also, urban areas are louder than rural areas (F = 20.49, P < 0.001). There is no interaction between visit and category (F = 1.16, P = 0.324).

Let’s look at date rather than visit.

## Robust linear mixed model fit by DAStau 
## Formula: mean_leq ~ Day * Category + (1 | Site) 
##    Data: dayavgl_20 
## 
## Scaled residuals: 
##    Min     1Q Median     3Q    Max 
## -2.973 -0.578  0.006  0.604  5.301 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Site     (Intercept) 3.73     1.93    
##  Residual             2.56     1.60    
## Number of obs: 268, groups: Site, 21
## 
## Fixed effects:
##                    Estimate Std. Error t value
## (Intercept)       -64.29337    1.46812   -43.8
## Day                 0.00449    0.00521     0.9
## CategoryRural      -6.73037    2.76203    -2.4
## Day:CategoryRural   0.00559    0.00966     0.6
## 
## Correlation of Fixed Effects:
##             (Intr) Day    CtgryR
## Day         -0.934              
## CategoryRrl -0.532  0.496       
## Dy:CtgryRrl  0.503 -0.539 -0.935
## 
## Robustness weights for the residuals: 
##  213 weights are ~= 1. The remaining 55 ones are summarized as
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   0.254   0.469   0.709   0.696   0.921   0.998 
## 
## Robustness weights for the random effects: 
##  17 weights are ~= 1. The remaining 4 ones are
##     2    10    11    15 
## 0.972 0.538 0.627 0.440 
## 
## Rho functions used for fitting:
##   Residuals:
##     eff: smoothed Huber (k = 1.345, s = 10) 
##     sig: smoothed Huber, Proposal 2 (k = 1.345, s = 10) 
##   Random Effects, variance component 1 (Site):
##     eff: smoothed Huber (k = 1.345, s = 10) 
##     vcp: smoothed Huber, Proposal 2 (k = 1.345, s = 10)

We performed lmer ( mean_leq ~ Day * Category + (1 | Site) , data = dayavgl_20 ) and found:

Chisq Df Pr(>Chisq)
Day 1.556 1 0.2122
Category 21.234 1 0.0000
Day:Category 1.813 1 0.1781

Let’s graph the results:

We see similar results by category (Chisq = 21.23, df = 4, 268, P < 0.001), but with no difference over time (Chisq = 1.56, P = 0.212, cond R\(^2\) = 0.75, marg R\(^2\) = 0.39) and no interaction (Chisq = 1.81, P = 0.178).

It seems like rural environments might be changing more over time. Let’s investigate whether harvest might play a role.

We performed lmer ( mean_leq ~ mean_harvest + (1 | Site) , data = dayavgl_20_rural ) and rlmerfound:

Chisq Df Pr(>Chisq)
mean_harvest 4.497 1 0.0339
Estimate Std..Error t.value P
(Intercept) -68.7209 0.6415 -107.128 0.0000
mean_harvest 0.0405 0.0216 1.874 0.0648

Let’s check the assumptions:

Let’s graph the results:

We used USDA data on week end percent harvest in 2020 for field crops in Nebraska. This gave details on oats, wheat, dry beans, sorghum, corn, and soybeans. We restricted this list to corn and soybeans, as these are the major crops grown and harvested in Lancaster County, Nebraska. We took the mean week end percent harvested of these two crops during the study season and compared these to the rural recorded vibratory noise levels. The week end percent harvested was positively correlated with the daily average Leq for rural sites (Chisq = 1.87, df = 2, 78, P = 0.065, cond R\(^2\) = 0.57, marg R\(^2\) = 0.02).

24 Hours

Here we assessed how vibratory noise levels change throughout the day. We graph the calculated mean and standard error. The grey areas represent nighttime. We added vertical dashed lines where vibratory noise peaked throughout the day, coinciding with rush hours. This provides further evidence that road noise likely represents a large component of vibratory noise. We also see what would likely be significant differences by category following the findings across season.

Let’s look by visit as well

Stats Table